Research Note

Institutional AI Tutor Pilot Framework

Evaluate a narrow learning or work task before evaluating a platform. Novel output is not a success measure.

Aug 4, 20262 min readBy Dalton Anderson
In this article

Institutional AI Tutor Pilot Framework

Pilot principle

Evaluate a narrow learning or work task before evaluating a platform. Novel output is not a success measure.

Pilot charter

AreaRequired record
JobOne low-risk learning or onboarding outcome
ParticipantsIncluded and excluded groups, consent, support, withdrawal
SourcesApproved manifest, rights, versions, conflicts, data classes
AccountProduct, edition, administrator, region, model, applicable terms
AccessIdentity, roles, sharing, export, public access, deletion
BaselineCurrent learning or task performance without the tool
FidelitySampled claim and citation audits
OutcomeComparable independent assessment or task result
AccessibilityParticipant testing, accommodations, barriers, alternatives
IntegrityPermitted assistance, attribution, assessment rules
IncidentsReporting, containment, correction, notification, recovery
ExitStop, revise, expand, export, delete, and ownership decision

Evidence

Vendor documentation can establish stated features and terms. It cannot establish the organization's actual configuration, legal compliance, accessibility, learning impact, source quality, or policy fit.

The organization must verify account type, administrative controls, sharing behavior, retention, connected services, feedback behavior, and data handling.

Use a subject expert to audit citations and claims. Use an educator or process owner to define and assess the outcome. Use privacy, security, accessibility, legal, copyright, labor, procurement, and records expertise as required by context.

Decision

Stop when the task is unsuitable, sources are unauthorized, participants cannot withdraw, access cannot be controlled, incidents cannot be handled, or outcomes cannot be measured.

Expand only after the organization has evidence for the named task and a clear owner for continued review.

Sources

Follow the evidence.

  1. blog.google: notebooklm new features december 2024blog.google
  2. blog.google: notebooklm audio overviewsblog.google
  3. daltonanderson.ghost.io: googles ai tutor the future of personalized learningdaltonanderson.ghost.io
  4. edu.google.com: ai notebooklmedu.google.com
  5. journals.sagepub.com: fulljournals.sagepub.com
  6. NIST Generative AI Profilenvlpubs.nist.gov
  7. open.spotify.com: 1YXy6yyC4u2mnqeARVB5qvopen.spotify.com
  8. studentprivacy.ed.gov: privacy and education technologystudentprivacy.ed.gov
  9. support.google.com: notebooklmsupport.google.com
  10. support.google.com: 16164461support.google.com
  11. support.google.com: 16213268support.google.com
  12. support.google.com: 16212820support.google.com
  13. support.google.com: 16179559support.google.com
  14. support.google.com: 16215270support.google.com
  15. support.google.com: 16322204support.google.com
  16. support.google.com: 17003757support.google.com
  17. support.google.com: 17004255support.google.com
  18. daltonanderson.net: googles ai tutor the future of personalized learningdaltonanderson.net
  19. NIST AI Risk Management Frameworknist.gov
  20. www2.ed.gov: ai reportwww2.ed.gov
  21. youtu.be: 6BwWKkZ7aeAyoutu.be

From this episode

Two useful next steps.

Evergreen · 1 min

What NotebookLM Does With Your Sources

Understand how NotebookLM imports sources, retrieves passages, generates answers and artifacts, shows citations, and changes data boundaries across accounts and sharing.

Guide · 1 min

How to Verify AI Answers and Citations

Audit an AI answer claim by claim by checking source identity, passage accuracy, support, context, inference, missing evidence, currency, authority, and final wording.

Return to the episode